Morphological computation-based control of a modular, pneumatically driven, soft robotic arm

The dynamics of soft robotic bodies are typically complex and exhibit nonlinearities and a high-dimensional state space. As a result, such systems are difficult to model and, therefore, hard to control. In this work, we use a model-free approach by employing the concept of morphological computation, which understands the complexity of the dynamics of such bodies as potential computational resources that can be exploited, for example, for control. The validity of this approach has been previously demonstrated in a number of simulations as well on a number of simple soft robotic platforms. However, this work takes the approach a significant step further by implementing it on a highly complex pneumatically driven robotic arm consisting of multiple modular segments, bringing the morphological computation-based control approach closer to real industrial applications. We demonstrate that various oval shaped end point trajectories can be learned and be reproduced consistently in a remarkably robust fashion. The presented morphological computation setup needs no model of the highly complex robot. Moreover, by exploiting the seemingly unbeneficial complex dynamics as a computational resource, the learning task to implement a nonlinear and dynamic control can be reduced to simple linear regression.